Mark is positioned incorrectly after forming ligature

[NorbertLindenberg]

I found this bug in Firefox 55.0.3 and reported it as Mozilla bug 1401870; @jfkthame suggested filing it here.

The attached font and HTML file demonstrate a bug that was found during development of a Burmese font. Burmese requires glyph reordering, ligature formation, and mark-to-base and mark-to-mark positioning.

In the test font, the marks u (U+102F) and anusvara (U+1036) are positioned relative to the base nna (U+100F) via anchor attachments. When tested by themselves, this works fine. Positioning of the anusvara also works fine when a medial ra (U+103C) is present. However, when a medial ra is present, the font forms a ligature from medial ra and u in order to position them together. The rule for this is

feature blws {
    script mym2;

    lookup raMedial_uMark_myanmar {
        lookupflag IgnoreBaseGlyphs UseMarkFilteringSet @below_base_myanmar;
        sub raMedial-myanmar uMark-myanmar by raMedial_uMark-myanmar;
    } raMedial_uMark_myanmar;
};

After the ligature is formed, the anusvara is no longer positioned correctly. It appears that the shaping engine loses track of nna as the base and positions relative to the medial ra + u ligature instead.

The same problem does not occur in Safari 11.0 on macOS 10.12.6.

Firefox bug.zip

[NorbertLindenberg]

More configuration info: This bug occurs in Firefox 56 and in Chrome 61.0.3163.100 on macOS 10.12.6; it does not occur in Safari 11.0 on the same macOS or in Edge 40.15063.0.0 on Windows 10 1703.

[behdad]

Your GDEF table does not classify the ligature as a mark. That's the bug. Please fix that and reopen if it still doesn't work. Also, "what does Windows do" is also a good question to ask always. I'm fairly sure it does what we do.

[NorbertLindenberg]

Why should the raMedial_uMark-myanmar (uni103C102F) ligature be classified as a mark? I don’t want it to attach to anything, and I do want it to keep its spacing behavior.

The last cluster in the HTML file is U+100F U+103C U+102F U+1036. After reordering, the glyph sequence is
uni103C uni100F uni102F uni1036
After the ligature is formed, the glyph sequence should be
uni103C102F uni100F uni1036
The mark anusvara (1036) follows the base nna (100F) and so should attach to it, no matter what the GDEF says about uni103C102F.

As noted above, the bug seen in Firefox and Chrome does not occur in Edge or in Safari.

Sorry about the name confusion – Glyphs likes to use user-friendly names in the app and the features file, but then produces AGL-compatible production names in the generated font file.

[NorbertLindenberg]

Please reopen this bug based on the comment above – GitHub doesn’t seem to let me do it.

[behdad]

Thanks Norbert. You are right. We did not notice the reordering before.